JP2010262648A - Automated method for alignment of document object - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To align a group of objects in a document based on the contents of the objects. <P>SOLUTION: The apparatus for alignment of objects in a document includes: a computer-readable memory 28 which stores an object alignment system; and a processor 26 which is communicable with the memory 28 and executes the object alignment system. The object alignment system includes: an object detection unit 60 for identifying first and second object to be aligned in the document; a profile generation unit 64 for generating a one dimensional guideline profile for the first object based on a detection of saliency for the first object; and an alignment generation unit 66 and an object reconfiguration unit 68 for executing alignment of the first and second objects based on the guideline profile. <P>COPYRIGHT: (C)2011,JPO&INPIT

Description

  This case relates to document layout. For example, there may be applications associated with automated systems and methods for aligning objects, taking into account the saliency within the object.

  Content alignment (arrangement) consists of organizing (organizing) various objects (text, graphics, pictures, photographic images, etc.) constituting a digital document. The purpose of this processing is generally to improve the readability and usability of the document by organizing the document objects in the page so that the contents (contents) are arranged in line with the natural scanning pattern of the human eye. To do well.

  Graphical designers decide where to place text objects and pictorial objects (photos, pictures, etc.) in the document, based on intuition and experience, in order to provide a comfortable arrangement for the viewer. If the text is associated with an image object, the text is placed, for example, between the upper and lower horizontal guidelines corresponding to the top and bottom edges of the image object to the right or left of the image object. . A graphical designer, for example, aligns text with a human face. Alternatively, in the case of an enlarged portrait, the professional designer aligns the text object, for example with respect to the eye or line of sight in the face.

  For many applications, such as variable data applications where either text or images in the document change depending on the recipient, such a manual method is not cost effective. I will. There are automated methods for such alignment, but those methods simply centered the text between two horizontal guidelines, ie aligned with the upper and lower guidelines. . Such methods are often inflexible and often result in the viewer not being aesthetically pleasing. For this reason, object alignment remains a process that is normally performed manually during document editing.

US Patent Application Publication No. 2010/0091330 (US Patent Application No. 12/250248) US Patent Application Publication No. 2009/0208118 US Patent Application Publication No. 2009/0034859 US Patent Application Publication No. 2008/0304740 US Patent Application Publication No. 20080304708 US Patent Application Publication No. 2008/0307422 US Pat. No. 7,400,761

Itti, et al., “A Model of Saliency-Based Visual Attention for Rapid Scene Analysis.” IEEE Transactions on Pattern Analysis and Machine Intelligence, 20 (11): 1254-1259 (1998) L. Itti and C. Koch, “Computational Modeling of Visual Attention,” Nature Reviews Neuroscience, 2 (3): 194-203 (2001)

  There is a need for content-based automated registration methods that can be used in applications such as digital publishing, where high value is given to the overall aesthetic quality.

  In one aspect, a method for aligning objects in a document comprises, for an input electronic document, identifying at least first and second objects to be aligned on a page of the document. Including. Based on the saliency detection for each of the first and second objects, a one-dimensional guideline profile is generated for one or both of the first and second objects. The first and second objects are aligned based on the guideline profile, thereby generating a corrected document and outputting the corrected document.

  The method further includes assigning at least the first object to one of a plurality of object classes each associated with an individual saliency detection method. The plurality of classes include a first class associated with the first saliency detection method and a second class associated with the second detection method.

  In the step of assigning at least the first object to one object class, the first and second objects may be assigned to individual object classes. This assignment may be performed using a classifier trained to classify the object group according to the class.

  The plurality of classes include, for example, at least one of an image object class, a text object class, and a graphics (graphic) element class.

  The first class is, for example, an image object class, and the second class is, for example, a text object class.

  When one of the first and second objects is assigned to an image object class, the image object may be classified into, for example, one of at least two categories. Here, the first of these categories is the human category, and the second is the non-human category.

  When the image object is classified into the human category, the detection method may detect at least one of a human face and eyes, and the guideline profile peaks at a position corresponding to the human face or eye. Has a value.

  When the image object is classified into the non-human category, the detection method may include a process of generating a saliency map.

  The guideline profile may be based on a one-dimensional average of saliency in the saliency map.

  In this method, a first guideline profile is generated for the previous first object, a second guideline profile is generated for the second object, and the first and second objects are connected to the first guideline profile. And alignment based on the second guideline profile.

  In this alignment process, the first and second guideline profiles may be convolved (convolution is obtained) to identify the peak, and the objects may be aligned based on the position of the peak.

  In the alignment process, the peak of the first guideline profile may be aligned with the peak of the second guideline profile.

  One of the first and second objects may be a fixed object, the other of the first and second objects may be a movable (floating) object, and alignment may be The movable object may be moved.

  The first object may be an image object, and the second object may be a text object.

  A computer program may be provided that includes code representing instructions that, when executed on a computer, cause the computer to perform the method described above.

  The computer device includes a memory storing a group of instructions for executing the above-described method, and a processor that can communicate with the memory and executes the group of instructions.

  In another aspect, an apparatus for aligning objects in a document comprises a computer readable memory that stores an object alignment system. The object registration system includes object detection means for identifying first and second objects to be registered in a document, at least a one-dimensional guideline profile for the first object, and a saliency for the first object. Profile generating means for generating based on the detection results of the above and positioning means for generating alignment results of the first and second objects based on the guideline profile. A processor capable of communicating with the memory executes an object alignment system.

  In another aspect, a computer-implemented method for aligning objects in a document includes identifying at least first and second objects to be aligned on a page of an electronic document; and Assigning to each object an object class selected from a plurality of object classes including an image object class and a text object class, and for each object to be aligned, based on the assigned object class Selecting a saliency detection method; generating a saliency expression by applying the selected saliency detection method to each object to be aligned; and applying the selected saliency detection method to the saliency expression for each object to be aligned. Based on one-dimensional Generating a modified profile by registering at least one of the first and second objects based on a corresponding one-dimensional guideline profile, respectively, Including a step of translating one of the second objects with respect to the other, and outputting the corrected document.

It is a figure which shows typically the document containing the image object and text object which are the object of alignment. 6 is a flowchart illustrating a method for aligning object groups in a document according to an aspect of an embodiment. It is a functional block diagram which shows the apparatus for position alignment of the object group in a document based on another side surface of embodiment. It is a figure which shows the example of a production | generation of the saliency expression about a text object. It is a figure which shows an example of an image object. FIG. 6 is a diagram illustrating a generation example of a saliency expression in the form of a saliency map when the image object of FIG. 5 is classified as a non-source photo (picture) image according to one aspect of the embodiment. FIG. 7 is a diagram for explaining an example of generating a guideline profile for the image object in FIG. 5 from the saliency map in FIG. 6. It is a figure for demonstrating the production | generation example of the guideline profile about the photograph (picture) image classified into the human. It is a figure for demonstrating the example which aligns a several object based on each guideline profile. It is a figure for demonstrating convolution of two guideline profiles (an image object and a text object). It is a figure which shows the example of the realignment with respect to the image object classified into the human of the text object.

  Exemplary embodiments relate to an apparatus, method and computer program for alignment of multiple objects in a document. The exemplary alignment method is comprehensive in the sense that it can be used for multiple document objects that are characterized by different content (content), visual aspects, and the like.

  This method uses a saliency measurement to determine the proper alignment between two objects, such as an image object and a text object. The detection of saliency can be viewed as a simulation or modeling of the human visual gaze mechanism. As will be appreciated, some portions of the image attract more attention from the human observer than others. The saliency is the “importance” or “attraction” of visual information in an image. The purpose of the saliency detection method is to assign a high saliency to each region in the image that is likely to be the main focus of general observer's attention. Many of these methods are based on biological visual models. These models are for evaluating which part of the image attracts visual attention. Implementation of these methods on a computer system falls into one of two main categories. One of the two categories seeks several related punctual positions, including a corner detector (Harris), a blob (blob) detector (Laplace), and a face. It is known as a site of interest (ie, key point) detector such as a detector. The other is to obtain a more continuous map of relevance such as a saliency map. However, there is also a mixing method that aims to provide a combination of the above keypoint method and a continuous method. This embodiment is not limited to a specific type of the saliency detection method. In one example, the saliency detection method is selected based on the type of object (text or image) and / or the evaluation (determination) of the content of the object. For example, the saliency map may be a probability map that takes into account the contents of the object. In each example shown in this specification, two (or more) saliency detection methods are used and the results are combined.

  Examples of the saliency detection method that can be used in this embodiment include those disclosed in the following documents. U.S. Patent Application Nos. 12 / 400,277, 12 / 033,434 (U.S. Patent Application Publication No. 2009/0208118), U.S. Patent Application Publication No. 2008/0304740, ibid. 2008/0304708, 2008/0304742, 2006/0093184, U.S. Patent 7,400,761, Itti, et al., "A Model of Saliency- Based Visual Attention for Rapid Scene Analysis. ”IEEE Transactions on Pattern Analysis and Machine Intelligence, 20 (11): 1254-1259 (1998); Hou, et al.,“ Saliency Detection: A Spectral. Residual Approach, ”ICVPR (2007 ); Gao, et al., “Bottom-up saliency is a discriminant process”, ICCV2007; Gao, et al., “The discriminant center-surround hypothesis for bottom-up saliency,” NIPS2007; Jones, et al. “Statistical Color Models with Application to Skin Detection, ”IJCV (46), No. 1, pp. 81-96 (January 2002); L. Itti and C. Koch,“ Computational Modeling of Visual Attention, ”Nature Reviews Neuroscience, 2 (3): 194- 203 (2001) (hereinafter referred to as “Itti and Koch”); Huang, et al., “A MPEG-7 Based Content-aware Album System for Consumer Photographs,” Bulletin of the College of Engineering, NTU, No. 90, pp. 3-24 (Feb. 2004); Liu, et al., “Learning to Detect A Salient Object,” CVPR 2007; and Z. Wang, B. Li, “A Two-Stage Approach to Saliency Detection in Images, ”In ICASSP (March / April 2008).

FIG. 1 schematically shows an example of a page of the electronic document 10. This document includes a template 12 that defines the relationship between the height and width of a digital document (or one page of the document) to be output by printing or screen display. The digital document 10 includes, for example, one or more pages or a video (moving image) (a series of images forming a series). Here, a text object may be inserted into each of the series of images. One or more objects 14 and 16 are arranged on the document template 12. When these objects 14 and 16 are printed, they form the same page in the document. Each of the objects 14, 16, etc. has a defined boundary (bounding box) with a defined height and width (respectively h ₁ , w ₁ and h ₂ , w ₂ ). The upper guide light lower guidelines (UG ₁ , LG ₁ , UG ₂ , LG ₂ ) for each object define the positions of the upper and lower ends of the object, respectively. In the case of the object 14, the guidelines UG ₁ and LG ₁ are spaced by a distance h _{1. In} the case of the object 16, the guidelines UG ₂ and LG ₂ are spaced by a distance h ₂ . Alternatively or additionally, left and right guidelines define the left and right edges of the object, and the left and right guidelines are spaced apart by distances w ₁ and w ₂ , respectively.

  One or more objects may be labeled as fixed objects, such as image object 14. “Fixed” means that the object cannot freely move in the horizontal direction or the vertical direction on the template 12. One or more objects, such as text object 16, are labeled as floating objects. This label can be translated on the template 12 in at least one dimension (direction), in this example, in the vertical dimension (ie, along the side of the template that defines the height of the template). It means that. However, the object may be freely movable in two or more dimensions. For example, it may be possible to move in a plurality of dimensions (directions) orthogonal to each other or in a middle direction between the horizontal direction and the vertical direction.

In the method exemplified here, the alignment of two or more objects such as the objects 14 and 16 in the document template 12 is performed in consideration of the saliency of one or both of the objects 14 and 16. The alignment is realized by parallel movement of the movable object. For example, in the case of vertical alignment, after alignment, one or both of the upper and lower guidelines of the lower object (object 16 in this example) is the higher object. It is located between the 14 upper guidelines and the lower guidelines. However, the exact location between the upper and lower guidelines UG ₁ and UG ₂ is a function of the saliency detected from one or both of these objects.

  The objects 14 and 16 for the document may already be arranged on the template without considering any spatial arrangement as part of the workflow process. For example, the upper end of the first object 14 may be automatically aligned with the upper end of the second object 16.

  The example object 41 is an image object. The image object is, for example, a photographic image that describes something in the real world. The graphical (graphics) object is, for example, a graph, a chart, a line drawing, and the like, and may be handled in the same way as an image object in an example. Both types of objects have visual features that can be detected by the saliency detection method. As will be appreciated, the image objects illustrated in the drawings (FIGS. 5, 8, and 9) are shown as line drawings, but these represent photographic images (eg, color photographic images), for example, for those images. The expression data shown in the drawing is required. The object 16 is a text object, and mainly includes a natural language text string generated using a character group from a predetermined character set and one or more selected fonts. A text object may have some salient features, such as magnified text parts and unusual text such as expressions, in some cases, but any “on” pixels are Can be considered equivalent.

  Objects 14 and 16 may be represented in any format, such as PDF, JPEG, GIF, JBIG, BMP, TIFF, or other common file formats used for images, as long as they are in an appropriate format. Often, this format may be converted to other suitable formats prior to processing. The example document shown in FIG. 1 has one text object 16 and one image object 14, but it will be appreciated that the technique of this embodiment is applicable to any number of each type of object. Is possible. Each object includes, for example, an array of pixels, and each pixel is associated with a color value, or in the case of a multicolor image, a set of color values for each color plate. The term “pixel” is used herein to denote “image element” and is used to emphasize two-dimensional image elements and three-dimensional image elements (in this case, the three-dimensional nature of the three-dimensional image pixels). , Sometimes referred to as voxel). The height h and width w of the object may be expressed by the number of pixels of the template. Similarly, the x and y coordinates of the object may be expressed by the number of pixels with reference to the upper side and the side side of the template.

  FIG. 4 shows an example of a method for aligning objects in a document. Processing of this method begins at S100, where a digital document containing document objects such as objects 14 and 16 is input.

  In S102, a document object (image and text (body)) is automatically detected. A fixed object and a movable object are specified.

  In S104, the detected objects are classified based on the contents of the objects themselves. In particular, the objects are classified according to the object type (image, text, etc.). You may make it categorize further in the kind.

  In S106, a saliency expression of each object, such as a saliency map or other expressions, is generated according to the type / category of the detected object. For example, various detection means (detectors) can be used based on the type / category of the object.

  In S108, at least one function (guideline profile) indicating one-dimensional (eg, vertical method) saliency of each object is generated based on the output of the saliency detector.

  In S110, the relative translation of at least the first of these objects relative to the second is calculated based on the guideline profile. This calculation is performed by minimizing the cost function obtained from the guideline profile or by aligning the positions of the maximum value portions of the guideline profile.

  In S112, re-alignment of the objects is executed by adjusting the relative positions of the objects based on the function. For example, the text object 16 is repositioned to an optimal position relative to the fixed image object 14.

  In S114, the corrected document is output.

  The method ends at S116.

  FIG. 3 is a diagram showing an example of an apparatus for aligning the objects 14 and 16 in the document, and this apparatus is used for executing the method exemplified above, for example. This apparatus may be embodied on an electronic processing device such as the computer 20 illustrated. In other examples, the electronic processing device 20 may be one or more special purpose computing devices such as a network server, an Internet-based server, a desktop computer, a notebook computer, a personal data assistant (PDA), a mobile phone, or the like. A general-purpose computing device may be included. The apparatus 20 includes an input unit 22, an output unit 24, a processor 26 such as a CPU, and a memory 28. Computer 20 is configured to implement object alignment system 30. This system 30 is hosted by the computer 20 for alignment of objects such as image objects and text objects in the original input document. The object alignment system 30 may be any form of software, hardware, or a combination of both. The example object alignment system 30 is stored in a computer readable memory 28 (eg, in a non-volatile portion of the computer memory 28) and performs the example method described above with reference to FIG. Instruction group is included. The components 22, 24, 26, 28 of the computer 20 may be communicatively connected to each other by a data / control bus 34. The input unit and output unit may be individual or combined, and may include, for example, an input port, a modem, a network connection, and the like.

  The computer 20 receives a digital document to be processed such as the document 10 via, for example, the input unit 22, stores the document 10 in a memory such as a nonvolatile portion of the computer memory 28, and the object alignment system 30 To be processed. The document 10 is deformed by the object registration system 30, for example, by rearranging the objects in the document. The computer 20 is configured to store and / or output a modified document 36 generated from the input document 10 by the object registration system 30. The corrected document 36 is output to another component (component) of a workflow processing system (not shown) that performs further processing on the corrected document, or to the image output devices 38 and 4. For example, the computer 20 includes, for example, a printer 38 or can communicate with the printer 38, and the printer 38 draws a document 38 on a print medium using a color material such as ink or toner. To do. Further, the computer 20 includes, for example, a display 40 such as an LCD screen for displaying the corrected document on the screen, or can communicate with the display 40.

  Document 10 may be input from any suitable data source 50. Such data sources 50 may include other components of workflow processing (which may be in the computer 20) or external data sources such as workstations, databases, scanners, memory storage devices (eg, disks, memory sticks). (Registered trademark)). When the data source 50 is an external data source, the data source 50 is connected to an appropriate input / output (I / O) such as a modem or a USB port via a wide area network such as a cable, a telephone line, a local area network, or the Internet. ) The connection unit 22 may be temporarily or permanently connected to the computer 20 so as to be communicable.

  For the computer 20, the processor 26 may be a central processing unit (CPU) of the computer 20. However, hardwired (wired) electronic circuits such as general purpose computers, special purpose computers, programmed microprocessors or microcontrollers, peripheral integrated circuit elements, ASICs and other integrated circuits, digital signal processors, discrete element circuits That is, it will also be understood that one or more logic circuits, programmable logic devices (PLD, PLA, FPGA, PAL, etc.) may be used. In general, any processor capable of implementing a finite state machine capable of implementing the flowchart shown in FIG. 2 can be used to implement the object alignment method described above.

  The memory 28 may be in the form of a plurality of individual memories, or may be in the form of a combination of them. As a substantial computer-readable storage medium, random access memory (RAM) may be used. , Read-only memory (ROM), magnetic disk, magnetic tape, optical disk, flash memory, holographic memory, or any suitable combination thereof.

  As shown in FIG. 3, the object alignment system 30 includes processing components 60, 62, 64, 66, 68 to perform the method of the embodiment.

  In particular, the object detection unit 60 detects document objects and detects whether the document objects are fixed or movable with respect to a specific dimension (direction). The saliency detection unit selection unit 62 identifies an appropriate saliency detection method. The selection unit 62 communicates with the object type classification unit 70 and the image type classification unit 72. The classification unit 70 assigns an object type to each object. The classification unit 72 assigns an image category to an object classified as a photo (picture) image. The two classification units 70 and 72 may include various object examples of different types and categories (these examples include those that belong to these types or categories (positive) and may not include cases (negative) in some cases. Learning may be performed using a learning set of (good). The saliency detection unit selection unit 62 selects an appropriate saliency detection method for each object based on the outputs of the classification units 70 and 72. For example, FIG. 3 shows three saliency detectors 74, 76, and 78 as an example. The selection unit 62 calls an appropriate one of the saliency detection units for each image, and generates a saliency expression for the object of interest. The profile generation unit 64 receives the saliency expression, and generates a guideline profile indicating the saliency for one dimension (for example, the vertical direction) of the object based on the saliency expression. Based on the guideline profile, the alignment generation unit 66 calculates the amount of translation of one or both of the objects to bring the objects into an optimal alignment state.

  The object rearrangement unit 68 performs object alignment. This is accomplished, for example, by storing new coordinates for one or both of these objects and outputting a modified document. The new coordinate information is stored, for example, as a tag such as an HTML tag or included in a file accompanying the document.

  Further details of this apparatus and method are described below.

<Document Object Detection: S102>
In S102, all objects constituting the document are identified, and the position of each object in the page is obtained. There are direct and indirect methods for detecting an object in a document. If the document is expressed in an open format such as XSL-FO (Extensible Stylesheet Language Formatting Objects: XSL is a language for formatting XML data), this step may be executed directly. For example, Apache FOP is an open source print formatter (typesetting device) that operates with an XSL formatting object (XSL-FO), and is an output-independent formatter. This application reads a formatting object (FO) tree and renders (draws) the resulting page group in a specified output format such as PDF, PS, PCL, AFP, or XML. Xiaofan Lin, Active Document Layout Synthesis, 8th International Conference on Document Analysis and Recognition, Seoul, Korea (2005); Itti and Koch, “A saliency-based search mechanism for overt and covert shifts of visual attention.” Vision Research, v. See also 40. 1489-1506 (2000).

  For other documents, indirect methods can be used. Examples of indirect methods are KY Wong, RG Casey, FM Wahl, “Document analysis system,” IBM Journal of Research and Development (1982); and George Nagy, “Twenty years of document image analysis in PAMI,” IEEE Transactions. on Pattern Analysis and Machine Intelligence, Vol. 22, No. 1, (January 2000). For example, the document may be binarized into a collection of white and black pixels, and each block of black pixels may be characterized by various features.

For each of the detected objects 14 and 16, the center-of-gravity positions c ₁ and c ₂ and the maximum height and maximum width (h ₁ , w ₁ and h ₂ , w _{2, respectively} ) are obtained (FIG. 1).

<Document classification: S104>
Each of the objects 14 and 16 is classified into one of a plurality of classes (classifications). In the embodiment, one main class is used. That is, text islands (a chunk of text), photographs (painting) images (such as continuous tone images, eg photographs), and graphics elements. For photographic images and graphics images, further classification steps are performed.

  In the evaluation of the object type, for example, the classification unit 70 is used to assign one class in the set of object classes to each object. Alternatively, a class may be assigned to an object based on a tag group associated with the object or document. In this case, in S104, a tag group is read to identify a class to be assigned to the object. In the embodiment, three classes of text, graphics, and image (photo / picture) objects are used, but other classifications may be used.

  Graphics (graphic) images can be classified using an automatic method such as a generic visual categorizer (GVC). This step is useful for understanding the content of the image and can be used as a priori information to better adjust the processing of the next step group.

  For objects classified as photo (picture) images, further investigation may be performed to evaluate the content of the images. For this step, one or a set of category classifiers 72 may be trained so that one of a plurality of content-based classes (categories) for a photo (picture) image can be assigned. . The category classification unit may learn in advance using a group of images that are manually labeled according to the class. For example, using a general visual classifier (GVC), it is possible to classify images into two or more different categories among people, faceless images, buildings, sky, seascapes, land landscapes, etc. it can. Examples of such classifiers are those disclosed in US Patent Application Publication Nos. 2007/0005356, 2007/0258648, and 2008/0069456. For example, there may be a simplified system that uses three categories (or simply uses a classifier other than person / person): sky, person, and non-human image.

  In this way, each object can be tagged with a label representing its own class (graphics, images (such as photos) or text), and in the case of image objects (such as photos) It can be tagged with a label representing a category (empty, person, or non-person) based on (content).

<Saliency detection: S106>
A distribution (guideline profile) is obtained for each detected object. The guideline profile indicates a position in a good vertical (vertical) direction as a guideline (reference line) for alignment. Consider three different ways to determine the distribution according to the label of the object.

  For text objects, the classification unit 74 generates a simple chi-square distribution expressed by parameters according to the dimensions (dimensions) of the object. For example, as shown in FIG. 4, the distribution 80 can be used as a guideline profile for the text object 14. This can be used to approximate a guideline profile for text. Alternatively, the geometric center (for example, the center of gravity) of a text box (a rectangular region that defines the range of text) may be used.

  For image objects classified as non-human (ie, for general images that do not include human faces), a traditional saliency map may be used (eg, Itti and Koch, “A saliency-based search mechanism for overt and covert shifts of visual attention. ”Vision Research, v. 40. 1489-1506 (2000)). For example, FIG. 5 shows an object 14 as an example, in this case, a photograph (picture) image that has been classified into categories other than people. FIG. 6 shows the saliency expression of the image forming the object 14 in the form of a saliency map 82. The image is divided into a plurality of small areas. In FIG. 6, for example, a lattice composed of 32 × 22 rectangular regions is shown. One saliency is assigned to each region, and this saliency is represented in the figure according to a gradual scale (for example, a set of gray scale values), where white has a high saliency, Black indicates low saliency. The used detection unit 74 gives high saliency to the regularly shaped regions on both sides of the Statue of Liberty. The guideline profile 84 is generated by averaging each row along the horizontal axis to obtain a vertical guideline profile (illustrated in FIG. 7). This type of saliency detection method can also be used for graphic (picture) images that have not been assigned to a graphics element or “person” class. Instead of the above example, a specially constructed saliency detector may be used.

  For image objects labeled with the human category, the Viola-Jones face detector (P. Viola and M. Jones. “Robust real-time face detection.” International Journal of Computer Vision (IJCV) 57 (2) 137 -154 (2004)) may be used. This method assumes that the face is the most prominent area in the image. For example, FIG. 8 shows an object 14 classified into a person category (a face is detected from a circle 86 of this object). The position of both eyes can be determined by simple calculation of the face size (indicated by line 88). Alternatively, more complex methods aimed at detecting both eye positions directly (eg GC Feng, PC Yuen, “Multi-cues eye detection on gray intensity image,” Pattern Recognition, See Elsevier (2001)). When the vertical position of both eyes is obtained, the center of the Gaussian distribution 90 is generated at the estimated height position of both eyes, and the Gaussian distribution can be displayed as a parameter relative to the face size. The Gaussian distribution can be used as a guideline}.

  In another example, a hybrid method (a method in which a plurality of methods are mixed) may be used. For example, two saliency maps of a saliency map based on face detection and a conventional saliency map are generated, and the two results are combined to obtain a combined saliency map used to generate a profile.

  In yet another example, saliency for all types of photographic (picture) images may be detected using the method described in the above-mentioned US patent application Ser. No. 12 / 400,277. Good. In this method, a target object is compared with an object (image) group in a set of a large number of stored objects (stored objects) to search for a set of stored objects having the highest similarity. Each stored object has a region of interest that is manually labeled. By using each region of interest of each retrieved object as a positive (ie, as a prominent example), the classifying unit is trained to classify each region in the object 14 as prominent or non-prominent, and is based on the content. Generate a degree map as a result.

<Estimation of guideline profile: S108>
As described above, for each of the objects 14 and 16, in order to facilitate the alignment of the two document objects, vertical (vertical) and horizontal (horizontal) alignment profiles 80, 84, 90 (guideline profiles). ) Is estimated. These profiles are distributions showing the likelihood that the alignment guidelines 94, 96 (FIG. 9) will be placed at specific vertical positions that reflect aesthetic and / or perceptual criteria. In FIG. 9, these alignment guidelines are arranged at the position of the local maximum value of the distribution. The profile is subjected to smoothing and other processing to facilitate calculation. For example, as shown in FIG. 7 (peaks P ₁ and P ₂ ), when there are two or more peaks in the distribution (profile), the smaller peak is omitted from the guideline profile 84 (FIG. 9).

  Instead, as illustrated in FIG. 10, a convolution 98 (eg, multiplication) of two guideline profiles (guidelines 80 and 84 in this example) is calculated, and the point where the maximum value is obtained is estimated as the realignment point 100. You may make it do. Here, the point of the maximum value of the maximum peak is regarded as an alignment guideline. In FIG. 10, the guideline profile and its convolution are shown not in the vertical direction but in the horizontal direction.

<Object re-alignment: S110, S112>
Based on the guideline profile, the objects are realigned, for example by minimizing the cost function.

  The re-alignment of the object is realized, for example, by shifting the position of one or both of the two objects with respect to the other party. When the position of one object is fixed as in the image object 14 of FIG. 9 and the other object 16 can be moved freely (ie, floating), the alignment of the movable (floating) object is performed. The position of the movable object is shifted so that the guideline matches the alignment guideline of the other object. This can be done very simply by calculating the vertical difference s between the maximum values of the two guideline profiles (ie the distance between the two alignment guidelines 94 and 96). Then, the movable object 16 is moved vertically by the distance s in the direction indicated by the arrow A, whereby the alignment guidelines 94 and 96 are aligned in the horizontal direction (that is, positioned on a straight line). The new position of the object 16 within the document page 10 is stored. Instead, as described above, the convolution 100 of the two guidelines may be calculated, and the point at which the convolution result becomes the maximum value may be used as the realignment point.

  In the example described above, the example in the case of the alignment of the image object and the text object has been particularly described. However, the method of this embodiment is applicable to a text object, an image object, a graphics object, a textual object, and those You may apply to the alignment of the 1st object and 2nd object which were each independently selected from the object which is a combination of these. Further, the first object may be aligned with respect to the second object in the vertical direction, and the first object may be aligned with respect to the third object in the horizontal direction. It is also conceivable to align the first object with respect to two or more other objects in the vertical and horizontal directions. For example, aligning the second object to the first peak of the alignment profile of the first object, aligning the third object to the second peak of the alignment profile of the first object, etc. . In yet another example, a horizontal alignment profile and a vertical alignment profile are generated for each of the first object and the second object, based on which matches best on the page. We propose to align two objects in either or both the horizontal and vertical directions. Although the boundaries of the illustrated objects 14 and 16 are separated from each other, in other examples, the movable first object overlaps in whole or in part with the second object (eg, in the latter). May be considered).

  The method illustrated in FIG. 2 may be implemented, for example, in a computer program product that is executed on a computer. The computer program product may be a tangible computer-readable recording medium recording a control program, such as a disk or a hard disk drive. Common forms of computer-readable media include floppy (registered trademark) disks, flexible disks, hard disks, magnetic tapes and other magnetic storage media, CD-ROMs, DVDs and other optical media, RAM, PROM, EPROM, flash There may be a picture PROM or other memory chip or memory cartridge, or other media that the computer can read and use. Alternatively, it may be in the form of a transmittable carrier wave representing a control program as a data signal transmission medium, such as an acoustic wave, a light wave, a wave generated in wireless data communication or infrared data communication.

  The methods exemplified above are hardwired (wired) such as general purpose computers, special purpose computers, programmed microprocessors or microcontrollers, peripheral integrated circuit elements, ASICs and other integrated circuits, digital signal processors, discrete element circuits. May be implemented using one or more of electronic circuits or logic circuits, programmable logic devices (PLD, PLA, FPGA, PAL, etc.). In general, any processor capable of implementing a finite state machine capable of implementing the flowchart shown in FIG. 2 can be used to implement the object alignment method described above.

<Application>
The systems and methods of the embodiments can be used for variable data (eg, variable data printing) applications such as one-to-one personalization (personal customization) and direct mail marketing. In the generation of a variable document, there are various problems in ensuring an appropriate aesthetic level for a portion that typically includes dynamically changing content (content). One challenge is how to dynamically handle the visual aspects of variable data workflows so that emphasis and management can be handled in a more contextual context. is there. The method described in the embodiment helps to cope with this by aligning document objects based on semantic content.

  For other applications such as image / document asset management and document image / photo set visualization, object alignment also benefits (e.g. photo post-processing).

  The following examples show application examples of the method of the embodiment, but these are not intended to limit the scope of the above-described embodiment.

<Example>
Using the automated prototype system developed as described above, the appropriate position of the movable text island 16 relative to the object 14 shown in FIG. 8 was calculated. The system correctly aligned the text island 16 to approximately the eye level of the person 14 in the object 14 in the vertical direction, as schematically illustrated in FIG. This is a more aesthetic result than the conventional system that aligns the top or bottom edges of the object.

  20 computers, 22 input units, 24 output units, 26 processors, 28 memories, 30 object alignment systems, 38 printers, 40 display screens, 50 document data sources, 60 object detection units, 62 saliency detection unit selection units, 64 profiles Generation unit, 66 Position generation unit, 68 Object rearrangement unit, 70 Object type classification unit, 72 Image object category classification unit, 74, 76, 78 Saliency detection unit

Claims (4)

A method for alignment of objects in an electronic document, comprising:
Identifying for the input electronic document a first object and a second object to be aligned within the page of the object;
Generating, by a computer processor, a one-dimensional guideline profile for at least the first object based on detection of saliency for the first object;
Aligning the first object and the second object based on the guideline profile to generate a modified document;
Outputting the modified document;
Including methods.   2. The method of claim 1, further comprising assigning at least the first object to one object class of a plurality of object classes, each object class being associated with one saliency detection method. The plurality of object classes include a first class associated with the first saliency detection method and a second class associated with the second saliency detection method. A method characterized by that.   The method according to claim 2, wherein when the image object is classified into a category representing a person, the saliency detection method includes a process of detecting a face and eye positions of at least one person, The guideline profile takes a peak value at a position corresponding to a human face or a human eye, and when the image object is classified into a category representing a person other than a human, the saliency detection method performs a process of generating a saliency map A method characterized by including. A device for aligning objects in a document,
A computer readable memory for storing the object alignment system;
A processor capable of communicating with the memory and executing the object alignment system;
With
The object alignment system includes:
Object detection means for identifying first and second objects to be aligned in the document;
Profile generation means for generating a one-dimensional guideline profile for at least the first object based on a saliency detection result for the first object;
Alignment means for performing alignment of the first and second objects based on the guideline profile;
A device comprising:

Images